KR20200085148A - Apparatus and method for performing wireless communication based on neural network model - Google Patents

Abstract

Disclosed are a device and method for performing wireless communication. According to an exemplary embodiment of the present disclosure, the device may include a transceiver and a controller connected to the transceiver. The controller may be configured to identify at least one added sample using a neural network model based on a digital signal and upscale the digital signal by adding at least one identified added sample to a plurality of samples of the digital signal.

Description

Device and method for performing wireless communication based on neural network model{APPARATUS AND METHOD FOR PERFORMING WIRELESS COMMUNICATION BASED ON NEURAL NETWORK MODEL}

The technical idea of the present disclosure relates to a device and a method of operating the same, and more particularly, to a device and a method of operating the wireless communication based on a neural network model.

A neural network refers to a computerized architecture modeling a biological brain. With the recent development of neural network technology, research into analyzing input data and extracting useful information using neural network devices in various types of electronic systems has been actively conducted.

In wireless communication, the quality of a signal can be determined according to the state of the network carrying the signal. In particular, wireless communication in a narrow-band network may frequently experience signal degradation due to loss of a signal frame or the like. In order to perform smooth communication regardless of the state of the wireless communication network, a technology capable of recovering the deteriorated signal to the original signal is required.

The technical idea of the present disclosure relates to an apparatus and method for performing wireless communication based on a neural network model, and provides an apparatus and method for upscailing a digital signal.

In order to achieve the above object, an apparatus for wireless communication according to an aspect of the technical idea of the present disclosure, the transceiver; As a controller connected to the transceiver, based on a digital signal, at least one additional sample is identified using a neural network model, and the identified at least one additional sample is added to a plurality of samples of the digital signal to add the digital signal. It includes a controller configured to upscaling (upscailing).

Meanwhile, a method for wireless communication according to another aspect of the technical spirit of the present disclosure includes: identifying at least one additional sample based on a digital signal using a neural network model; And upscailing the digital signal by adding the identified at least one additional sample to a plurality of samples of the digital signal.

Meanwhile, according to another aspect of the technical spirit of the present disclosure, a computer-readable recording medium in which a program for implementing a method of operating a controller is recorded is provided.

According to the present disclosure, an apparatus and method capable of upscaling a digital signal based on a neural network model can be provided. In addition, by generating a neural network model based on a reference digital signal, it is possible to provide an apparatus and method capable of obtaining a digital signal with improved signal quality.

1 shows a wireless communication system according to an exemplary embodiment of the present disclosure.
2 shows a flow chart for the operation of an apparatus according to an exemplary embodiment of the present disclosure.
3 shows an embodiment in which a device according to an exemplary embodiment of the present disclosure upscales a digital signal.
4 shows a flow chart for the operation of the device according to an exemplary embodiment of the present disclosure.
5 is a diagram illustrating an example of generating a neural network model according to an exemplary embodiment of the present disclosure.
6 illustrates an embodiment in which a device according to an exemplary embodiment of the present disclosure generates a neural network model.
7 is a diagram illustrating an example of generating a neural network model according to an exemplary embodiment of the present disclosure.
8 illustrates an embodiment in which a device according to an exemplary embodiment of the present disclosure generates a neural network model.
9 is a block diagram of an apparatus according to an exemplary embodiment of the present disclosure.
10 is a diagram illustrating an example of a network simulation according to an exemplary embodiment of the present disclosure.
11 is a diagram illustrating an example of a neural network model according to an exemplary embodiment of the present disclosure.
12 illustrates an embodiment of upscaling a digital signal based on a neural network model according to an exemplary embodiment of the present disclosure.
13 is a block diagram showing a configuration of an apparatus according to an exemplary embodiment of the present disclosure.
14 shows a wireless communication system according to an exemplary embodiment of the present disclosure.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 shows a wireless communication system according to an exemplary embodiment of the present disclosure.

Referring to FIG. 1, the wireless communication system may include a transmitting terminal 1, a base station 5, and a receiving terminal 10.

The base station 5 may communicate wirelessly with the terminals 1 and 10 through one or more base station antennas. The wireless communication network supported by the base station 5 can support multiple users communicating by sharing available network resources. For example, in a wireless communication network, GSM (Global System for Mobile communication), Code Division Multiple Access (CDMA), Frequency Division MultipleAccess (FDMA), Time Division MultipleAccess (TDMA), Orthogonal Frequency Division MultipleAccess (OFDMA), SC-FDMA ( Information may be transmitted in various ways such as Single Carrier-Frequency Division MultipleAccess.

The wireless communication network supported by the base station 5 can transmit voice signals. For example, in a wireless communication network, a voice signal may be transmitted in various ways, such as Global System for Mobile Communication (GSM) and Voice over Long Term Evolution (Volte). In the GSM (Global System for Mobile communication) scheme, the bandwidth may be limited to 4 kHz.

The voice signals exchanged between the terminals 1 and 10 may be compressed using an AMR (Adaptive Multi-Rate) audio codec, and may be encoded or decoded.

In this figure, one base station 5 is shown, but this is only for convenience of description, and the wireless communication system includes various numbers of base stations (eg, macro, micro, and/or pico). ) Base station).

The base station 5 can provide communication coverage for a given geographic area. In some examples, the base station 5 is a base transceiver station (BTS), a radio base station (radio base station), an AP (Access Point), a radio transceiver (radio transceiver), NodeB, eNodeB (eNB) or other It can be named in an appropriate term.

The terminals 1 and 10 are wireless communication devices, which may be fixed or mobile, and may refer to various devices that can communicate with the base station 5 to transmit and receive data and/or control information. For example, the terminals 1 and 10 are terminal equipment, mobile station (MS), mobile terminal (MT), user terminal (UT), subscriber station (SS), wireless device, mobile It may be referred to as a handheld device.

The transmitting terminal 1 may receive an analog type voice signal and convert it into a digital signal. The transmitting terminal 1 may transmit the voice signal converted into a digital signal to the receiving terminal 10 through the base station 5. The receiving terminal 10 may receive a digital audio signal and convert it into an analog signal. The receiving terminal 10 may output a voice signal converted into an analog signal through a speaker built in the receiving terminal 10.

In an exemplary embodiment, the terminals 1 and 10 may modulate an analog signal into a digital signal and modulate the digital signal into an analog signal through a PCM (Pulse Code Modulation) method.

2 shows a flow chart for the operation of an apparatus according to an exemplary embodiment of the present disclosure.

Referring to FIG. 2, the device 10 may identify at least one additional sample (S300). In an exemplary embodiment, the device 10 may identify at least one additional sample based on the input digital signal. In another exemplary embodiment, the device 10 may identify at least one additional sample based on the input digital signal using a neural network model. For example, the at least one additional sample may be a sample that does not correspond to a plurality of samples of the input digital signal.

Next, the device 10 may upscailing the digital signal (S400). In an exemplary embodiment, the device 10 may upscale the digital signal by adding at least one additional sample identified to the plurality of samples of the input digital signal. For example, the upscaled digital signal may have a higher sampling rate than the input digital signal.

3 shows an embodiment in which a device according to an exemplary embodiment of the present disclosure upscales a digital signal.

Referring to FIG. 3, the device 10 may upscale the input voice signal I_DIG to output the output voice signal O_DIG. For example, the input audio signal and the output audio signal are digital signals.

In an exemplary embodiment, the device 10 may receive and decode the encoded input voice signal I_DIG. For example, encoding and decoding may be based on the AMR method. In another exemplary embodiment, the decoded input speech signal I_DIG may be divided into a frame having a sampling rate of 8 kHz, a quantization level of 13 bits, and a length of 20 ms.

In another exemplary embodiment, the output audio signal O_DIG may have a higher sampling rate than the input audio signal I_DIG. For example, the output voice signal O_DIG may be upsampled n times, and may have a sampling rate of 32kHz in response to the 8kHz input voice signal I_DIG. In another exemplary embodiment, the output speech signal O_DIG may have a frame having the same length as the input speech signal I_DIG, and the output speech signal O_DIG has a higher quantization level than the input speech signal I_DIG. Can have

4 shows a flow chart for the operation of the device according to an exemplary embodiment of the present disclosure. Among the components shown in FIG. 4, descriptions of components overlapping with FIG. 2 will be omitted.

Referring to FIG. 4, the device 10 may generate a neural network model (S100). In an exemplary embodiment, the apparatus 10 may train the neural network model by inputting a digital signal into the neural network model and feeding back information related to the digital signal output from the neural network model to the neural network model.

In another exemplary embodiment, the device 10 inputs a pre-processed reference digital signal into the neural network model, and a digital signal output from the neural network model and a non-pre-processed reference digital signal. By feeding back the difference between the neural network models, the neural network model can be trained.

In another exemplary embodiment, the device 10 may include a neural network model already generated through learning. For example, the neural network model may be stored in a memory built into the device 10. If necessary, the device 10 can read the stored neural network model from memory.

Next, the device 10 may determine a weight based on the input digital signal (S200). In an exemplary embodiment, the device 10 may determine a weight using a neural network model. For example, the neural network model may modify the weight so that the digital signal output from the neural network model is similar to the pre-processed reference digital signal in response to the pre-processed reference digital signal.

Next, the device 10 may identify at least one additional sample (S300). In an exemplary embodiment, the device 10 may identify at least one additional sample based on the input digital signal and the determined weight. For example, the at least one additional sample may be a sample that does not correspond to a plurality of samples of the input digital signal.

5 is a diagram illustrating an example of generating a neural network model according to an exemplary embodiment of the present disclosure.

Referring to FIG. 5, the device 10 may acquire a first input digital signal (S110). In an exemplary embodiment, the device 10 may upscale the first input digital signal using a neural network model.

In another exemplary embodiment, the device 10 inputs the first input digital signal into the neural network model, and the first input digital based at least one additional sample identified from the neural network model based on the first input digital signal. The first input digital signal can be upscaled by adding it to the signal. For example, the neural network model may identify at least one additional sample based on the first input digital signal and a predetermined weight.

Next, the device 10 may determine whether there is a difference between a first output digital signal upscaled from the first input digital signal and a reference digital signal (S120). In an exemplary embodiment, the difference may be related to at least one sample that does not correspond to a plurality of samples of the first output digital signal among a plurality of samples of the reference digital signal.

If there is a difference, the device 10 may derive the difference (S130). In an exemplary embodiment, the device 10 is configured to input the derived difference and the second input digital signal into the neural network model and add at least one additional sample identified from the neural network model to the second input digital signal. You can upscale 2 input digital signals. For example, the neural network model can identify at least one additional sample based on the derived difference, the second input digital signal, and a predetermined weight. In another exemplary embodiment, the neural network model may modify the weight so that the difference between the reference digital signal and the second output digital signal does not occur based on the input of the derived difference.

If there is no difference, the device 10 may obtain a first output digital signal upscaled from the first input digital signal as an output digital signal (S140). In an exemplary embodiment, the device 10 may obtain the weight applied in the neural network model when there is no difference. For example, the device 10 may provide the neural network model to apply the obtained weights when upscaling the next input digital signal by the neural network model.

6 illustrates an embodiment in which a device according to an exemplary embodiment of the present disclosure generates a neural network model.

Referring to FIG. 6, the device 10 may include a processor 11 that upscales the input digital signal I_SIG. In an exemplary embodiment, the processor 11 may include a neural network model.

The processor 11 may upscale the input digital signal I_SIG to generate the output digital signal O_SIG. In an exemplary embodiment, the processor 11 may input an input digital signal I_SIG as a neural network model and obtain an output digital signal O_SIG from the neural network model. For example, the neural network model may determine the output digital signal O_SIG based on the input digital signal I_SIG and a predetermined weight.

The device 10 may include an operator 12 that obtains the difference between the output digital signal O_SIG and the reference digital signal S_SIG. For example, the reference digital signal S_SIG may be one of a set of reference digital signals including a plurality of reference digital signals, and the reference digital signal set may include voice signals of various languages and voice signals of various user samples. have.

The operator 12 may derive, as a difference, at least one sample that does not correspond to a plurality of samples of the output digital signal O_SIG from among a plurality of samples of one reference digital signal S_SIG. The calculator 12 can feed the difference difference back to the processor 11.

In an exemplary embodiment, the processor 11 is upscaled in the neural network model so that a difference between the reference digital signal S_SIG and the output digital signal O_SIG does not occur based on the difference received from the operator 12. You can modify the weight applied to the city.

In another exemplary embodiment, the processor 11 may input an input digital signal I_SIG and a difference received from the operator 12 as a neural network model, and obtain an output digital signal O_SIG from the neural network model. have. The neural network model may determine the output digital signal O_SIG based on the input digital signal I_SIG, a predetermined weight, and a difference received from the operator 12.

7 is a diagram illustrating an example of generating a neural network model according to an exemplary embodiment of the present disclosure. Among the configurations shown in FIG. 7, descriptions of configurations overlapping with FIG. 5 will be omitted.

Referring to FIG. 7, the device 10 may pre-process the reference digital signal (S105). In an exemplary embodiment, the process of pre-processing the reference digital signal encodes the reference digital signal, adds noise to the encoded reference digital signal, and decodes the encoded reference digital signal to which the noise is added. It may be a (decoding) process. The order of encoding, noise addition and decoding for pre-processing the reference digital signal may be different.

In another exemplary embodiment, the process of pre-processing a reference digital signal may correspond to a process of simulating signal transmission through a wireless communication network.

Next, the device 10 may acquire a pre-processed reference digital signal as a first input digital signal (S110).

Next, the device 10 may determine whether there is a difference between a first output digital signal upscaled from the first input digital signal and a reference digital signal that is not pre-processed (S120 ). In an exemplary embodiment, the difference may be related to at least one sample that does not correspond to a plurality of samples of the first output digital signal among a plurality of samples of the pre-processed reference digital signal.

If there is a difference, the device 10 may derive the difference (S130). In an exemplary embodiment, the device 10 inputs a second input digital signal pre-processed with a reference digital signal and a derived difference into a neural network model, and provides at least one additional sample identified from the neural network model. The second input digital signal can be upscaled by adding it to the two input digital signal. For example, the neural network model may identify at least one additional sample based on the derived difference, the second input digital signal pre-processing the reference digital signal, and a predetermined weight. In another exemplary embodiment, the neural network model may modify the weight so that the difference between the reference digital signal and the second input digital signal pre-processing the reference digital signal does not occur based on the input of the derived difference.

8 illustrates an embodiment in which a device according to an exemplary embodiment of the present disclosure generates a neural network model. Among the components shown in FIG. 8, descriptions of components overlapping with FIG. 6 will be omitted.

Referring to FIG. 8, the device 10 may further include a simulation module 13 that simulates transmission of a signal in a network. In an exemplary embodiment, the simulation module 13 may include an encoder 13c, a noise generator 13b, and a decoder 13a.

The simulation module 13 may generate an input digital signal I_SIG by pre-processing the input reference digital signal S_SIG. In the exemplary embodiment, the encoder 13c encodes the reference digital signal S_SIG, the noise generator 13b adds noise to the encoded reference digital signal, and the decoder 13a has noise. The added encoded reference digital signal can be decoded. For example, the operation order of the encoder 13c, the noise generator 13b, and the decoder 13a for the reference digital signal S_SIG may vary.

The processor 11 may generate an output digital signal O_SIG by upscaling the input digital signal I_SIG on which the reference digital signal is pre-processed.

The operator 12 may derive, as a difference, at least one sample that does not correspond to a plurality of samples of the output digital signal O_SIG from among a plurality of samples of the pre-processed reference digital signal S_SIG. The calculator 12 can feed the difference difference back to the processor 11.

In the exemplary embodiment, the processor 11 is based on the difference received from the operator 12, so that the difference between the pre-processed reference digital signal S_SIG and the output digital signal O_SIG does not occur. You can modify the weight applied when upscaling in the model.

In another exemplary embodiment, the processor 11 inputs the input digital signal I_SIG pre-processed with the reference digital signal S_SIG and the difference received from the operator 12 as a neural network model, and the neural network model. From it, an output digital signal (O_SIG) can be obtained. The neural network model may determine the output digital signal O_SIG based on the input digital signal I_SIG pre-processed with the reference digital signal S_SIG, a predetermined weight, and a difference received from the operator 12.

9 is a block diagram of an apparatus according to an exemplary embodiment of the present disclosure.

Referring to FIG. 9, the device 10 may include a controller 20 and a transceiver 30. In addition, although not shown, the device 10 may additionally include various components such as a memory.

In an exemplary embodiment, the memory is a program (or software), and may store a kernel, middleware, application programming interface (API), and application. At least a portion of the kernel, middleware, and API may be referred to as an operating system. The kernel, for example, can control or manage system resources (eg, a controller or memory) used to execute operations or functions implemented in a program. The middleware may act as an intermediary so that, for example, an API or application communicates with the kernel to exchange data. The application may be implemented to perform various upscailing effects on signals acquired through the transceiver 30, for example.

The transceiver 30 may include a communication module such as an antenna. In an exemplary embodiment, the transceiver 30 may receive digital signals transmitted through a wireless communication network. For example, the transceiver 30 may transmit digital signals taken in the time domain to the controller 20.

The controller 20 may include one or more processors. In an exemplary embodiment, the controller 20 may control the transceiver 30 to receive a digital signal. In another exemplary embodiment, the controller 20 may receive a digital signal in the time band from the transceiver 30, modulate the digital signal in the time band, and output an analog signal in the time band.

In another exemplary embodiment, the controller 20 is a single processor, and performs all control operations performed by the device 10, such as operations performed by the processor 11, the operator 12, and the simulation module 13 can do.

In addition, although not shown, the controller 20 may include a processor 11, an operator 12 and a simulation module 13. The processor 11 may, for example, upscale the input digital signal I_SIG to generate the output digital signal O_SIG. The operator 12 may derive a difference, for example, by comparing one reference digital signal S_SIG and an output digital signal O_SIG. The simulation module 13 can, for example, pre-process the reference digital signal.

Further, the processor 11 may input, for example, an input digital signal I_SIG as a neural network model, and obtain an output digital signal O_SIG from the neural network model. For example, the neural network model may be driven through one or more computing devices of a central processing unit (CPU), a graphics processing unit (GPU), and a neural processing unit (NPU).

10 is a diagram illustrating an example of a network simulation according to an exemplary embodiment of the present disclosure.

Referring to FIG. 10, one reference digital signal of one of a dataset including one or more reference digital signals (Ground Truth audio) passes through the encoder 14 and the decoder 15, and then is neural as an input digital signal. It can be entered as a network model. In an exemplary embodiment, the reference digital signal can have a sampling rate of 32 kHz. In another exemplary embodiment, the reference digital signal may be divided into frames having a length of 20 ms.

The reference digital signal may be input to the encoder 14 and encoded. The encoded reference digital signal may be input to a decoder 15 and decoded. In an exemplary embodiment, the process of encoding and decoding a reference digital signal may correspond to a process of simulating signal transmission through a wireless communication network. In another exemplary embodiment, the reference digital signal may be encoded and decoded for each frame having a predetermined time length. For example, the time length of the frame may be 20 ms. In another exemplary embodiment, the encoder 14 and the decoder 15 may use an AMR (Adqptive Multi-Rate) method.

The reference digital signal passing through the encoder 14 and the decoder 15 may be input to a neural network as an input digital signal. In an exemplary embodiment, the input digital signal may be downscaled than the reference digital signal. For example, the input digital signal can have a sampling rate of 8 kHz. In another exemplary embodiment, the input digital signal may be divided into frames having the same time length as the reference digital signal. For example, the input digital signal may consist of frames having a length of 20 ms.

11 is a diagram illustrating an example of a neural network model according to an exemplary embodiment of the present disclosure.

Referring to FIG. 11, the neural network NN may have a structure including an input layer, hidden layers, and output layers. The neural network NN may perform an operation based on the received input data (eg, I1 or I2) and generate output data (eg, Q1 or Q2) based on the result of the execution. In an exemplary embodiment, the neural network NN is employed in the device 10 to receive the input digital signal I_SIG and generate an output digital signal O_SIG close to the reference digital signal S_SIG.

The neural network NN may be a deep neural network (DNN) or n-layers neural networks including two or more hidden layers. For example, as illustrated in FIG. 11, the neural network NN may be a DNN including an input layer 210, first and second hidden layers 212 and 213, and an output layer 216. DNN may include, but is not limited to, Convolution Neural Networks (CNN), Recurrent Neural Networks (RNN), Deep Belief Networks, Restricted Boltzman Machines, and the like.

When the neural network NN has a DNN structure, since the neural network NN includes more layers capable of extracting valid information, the neural network NN can process complex data sets. Meanwhile, the neural network NN is illustrated as including four layers 210, 212, 214, and 216, but this is only an example, and the neural network NN may include fewer or more layers. . In addition, the neural network NN may include layers having various structures different from those shown in FIG. 11.

Each of the layers 210, 212, 214, and 216 included in the neural network NN may include a plurality of neurons. Neurons may correspond to a plurality of artificial nodes, known as processing elements (PEs), units, or similar terms. For example, as illustrated in FIG. 11, the input layer 210 has two neurons (nodes), and each of the first and second hidden layers 212 and 214 has three neurons (nodes). It can contain. However, this is only an example and each of the layers included in the neuron network NN may include various numbers of neurons (nodes).

Neurons included in each of the layers included in the neuron network NN may be connected to each other to exchange data. One neuron can receive data from other neurons and calculate them, and output the result of the calculation to other neurons.

The input and output of each of the neurons (nodes) may be referred to as input activation and output activation. That is, the activation may be an output of one neuron and may be a parameter corresponding to the input of neurons included in the next layer. On the other hand, each of the neurons may determine its own activation based on the weights and the activations received from the neurons included in the previous layer. The weight is a parameter used to calculate the output activation in each neuron, and may be a value assigned to a connection relationship between neurons.

는 액티베이션 함수(activation function)이고,

는 (i-1) 번째 레이어에 포함된 k 번째 뉴런으로부터 i 번째 레이어에 포함된 j 번째 뉴런으로의 가중치 값일 수 있다.

는 i 번째 레이어에 포함된 j 번째 뉴런의 바이어스(bias) 값이고,

는 i 번째 레이어의 j 번째 뉴런의 액티베이션, 다시 말해서 포스트 액티베이션(post activation)으로 지칭될 수 있다. 포스트 액티베이션

는 다음의 [수학식 1]을 이용하여 계산 될 수 있다.Each of the neurons can be processed by a computational unit or processing element that receives input and outputs activation, and the input-output of each of the neurons can be mapped. For example

Is the activation function,

May be a weight value from the k th neuron included in the (i-1) th layer to the j th neuron included in the i th layer.

Is a bias value of the j-th neuron included in the i-th layer,

May be referred to as activation of the j-th neuron of the i-th layer, that is, post activation. Post activation

Can be calculated using the following [Equation 1].

로 표현될 수 있다. 또한,

은 [수학식 1]에 따라

의 값을 가질 수 있다. 다시 말해서, 포스트 액티베이션은 이전 레이어로부터 수신된 액티베이션들의 합(sum)에 액티베이션 함수를 적용하여 획득된 값일 수 있다. 다만, [수학식 1]은 뉴럴 네트워크에서 데이터를 처리하기 위해 이용되는 액티베이션 및 가중치를 설명하기 위한 예시일 뿐, 이에 제한되지 않는다.As shown in FIG. 11, the post activation of the first neuron of the first hidden layer 212 is

Can be expressed as Also,

According to [Equation 1]

It can have the value of In other words, the post activation may be a value obtained by applying an activation function to the sum of activations received from the previous layer. However, [Equation 1] is only an example for describing activation and weights used to process data in a neural network, but is not limited thereto.

12 illustrates an embodiment of upscaling a digital signal based on a neural network model according to an exemplary embodiment of the present disclosure.

Referring to FIG. 12, the input digital signal I_SIG is input to the neural network model and processed in each layer L_VAR, so that it can be upscaled to the output digital signal O_SIG. In an exemplary embodiment, each layer (L_VAR) of the neural network model may model an encoder and a decoder.

In another exemplary embodiment, layers L_VAR on the encoder side may be concatenated with layers L_VAR on the decoder side. For example, in the case of the downsampling layer L_VAR on the encoder side, skip connections to the layer L_VAR on the decoder side may be added.

In another exemplary embodiment, each layer L_VAR of the neural network model may be generated by learning from a digital signal.

In another exemplary embodiment, the loss function of the neural network may be calculated using [Equation 2].

13 is a block diagram showing a configuration of an apparatus according to an exemplary embodiment of the present disclosure.

Referring to FIG. 13, operations of the present invention may include operations performed at the device 10 level and operations performed at the controller 20 level. As an operation performed at the device 10 level, a digital signal from a cellular network may be received by a baseband processor. In addition, the digital signal output from the controller 20 may be output as a speaker output. To this end, the device 10 may modulate, for example, a digital signal into an analog signal and output it to a speaker.

As an operation performed at the controller 20 level, the controller 20 may access the digital signal through an audio hardware abstraction layer and introduce the digital signal into the AudioFlinger / AudioMixer. The AudioFilnger/ AudioMixer can transmit the incoming digital signal to the MediaRecorder, and the MediaRecorder can input the digital signal modulated with the narrow band into the Deep Neural Network. The Deep Neural Network can receive and process digital signals modulated in a narrow band to generate digital signals modulated in a wide band. The broadband modulated digital signal output from the Deep Neural Network can be returned to the AudioFlinger / AudioMixer for output on the speaker.

14 shows a wireless communication system according to an exemplary embodiment of the present disclosure.

Referring to FIG. 14, the wireless communication system may include a transmitting side terminal 1b and a receiving side terminal 10b. In an exemplary embodiment, the transmitting terminal 1b may receive an analog-type voice signal and convert it into a digital signal. Signals exchanged between the terminals 1b and 10b may not require a base station. The receiving terminal 10b may receive a digital type voice signal and convert it into an analog signal. The receiving terminal 10b may output a voice signal converted into an analog signal through a speaker built in the receiving terminal 10b.

As described above, exemplary embodiments have been disclosed in the drawings and the specification. Although the embodiments have been described using specific terminology in this specification, they are used only for the purpose of describing the technical spirit of the present disclosure and are not used to limit the scope of the present disclosure as defined in the claims or the claims. . Therefore, those of ordinary skill in the art will understand that various modifications and other equivalent embodiments are possible therefrom. Therefore, the true technical protection scope of the present disclosure should be defined by the technical spirit of the appended claims.

Claims (15)

A device for wireless communication,
Transceiver; And
And a controller connected to the transceiver,
The controller:
At least one additional sample is identified based on the digital signal using a neural network model,
And configured to upscailing the digital signal by adding the identified at least one additional sample to a plurality of samples of the digital signal. According to claim 1,
The controller uses the neural network model to identify the at least one additional sample based on the digital signal,
Determine a weight in response to the digital signal,
And an apparatus configured to identify the at least one additional sample based on the digital signal and the weight. According to claim 1,
The controller,
Obtaining an upscaled first output digital signal from the first input digital signal in response to a first input digital signal, and the difference between one reference digital signal and the first output digital signal from at least one set of reference digital signals And generating the neural network model through acquiring a second input digital signal and an upscaled second output digital signal from the second input digital signal based on the difference. According to claim 3,
And the difference is related to at least one sample among the plurality of samples of the one reference digital signal that does not correspond to the plurality of samples of the first output digital signal. According to claim 3,
The controller,
A device configured to generate the first input digital signal and the second input digital signal by pre-processing a first reference digital signal and a second reference digital signal, respectively, among the at least one reference digital signal. . The method of claim 5,
The controller pre-processes the first reference digital signal and the second reference digital signal, respectively,
The first reference digital signal and the second reference digital signal are respectively coded, noise is added to the coded first reference digital signal and the coded second reference digital signal, and the coded first is added with the noise. And a reference digital signal and a coded second reference digital signal to which the noise is added. The method of claim 5,
The sampling rate of the first reference digital signal and the sampling rate of the second reference digital signal are higher than the sampling rate of the first input digital signal and the sampling rate of the second input digital signal, respectively. As a method for wireless communication,
Identifying at least one additional sample based on the digital signal using a neural network model; And
And upscailing the digital signal by adding the identified at least one additional sample to a plurality of samples of the digital signal. The method of claim 8,
The step of identifying using the neural network model,
Determining a weight in response to the digital signal; And
And identifying the at least one additional sample based on the digital signal and the weight. The method of claim 8,
Further comprising the step of generating the neural network model,
Generating the neural network model,
Obtaining an upscaled first output digital signal from the first input digital signal in response to a first input digital signal;
Obtaining a difference between one reference digital signal and the first output digital signal from at least one set of reference digital signals; And
And obtaining an upscaled second output digital signal from the second input digital signal based on the second input digital signal and the difference. The method of claim 10,
Wherein the difference is related to at least one sample that does not correspond to a plurality of samples of the first output digital signal among a plurality of samples of the one reference digital signal. The method of claim 10,
Pre-processing a first reference digital signal and a second reference digital signal among the at least one reference digital signals, respectively, to generate the first input digital signal and the second input digital signal. How to include. The method of claim 12,
Generating the first input digital signal and the second input digital signal,
Encoding the first reference digital signal and the second reference digital signal, respectively;
Adding noise to the coded first reference digital signal and the coded second reference digital signal; And
And decoding the coded first reference digital signal to which the noise is added and the coded second reference digital signal to which the noise is added, respectively. The method of claim 12,
The sampling rate of the first reference digital signal and the sampling rate of the second reference digital signal are higher than the sampling rate of the first input digital signal and the sampling rate of the second input digital signal, respectively. A computer-readable recording medium recording a program for executing the method of any one of claims 8 to 14 on a computer.

Images